CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation
occlusion areas lead to mismatches during the image matching
algorithm and subsequently to errors in the resulting surface
model.
3. DATA SET AND STUDY AREA
The satellite Ikonos is able to rotate the CCD Linear Array
sensor up to an angle of 26° off-nadir, so the satellite can take
images of the same location from two different view points on
the same orbital track. Next to along track stereo pairs, it is also
possible to create stereo couples out of images from the same
area but taken from a different orbit at a different date. These
are so-called across track stereo pairs. This approach to form
couples has some disadvantages. The most important ones are
radiometric differences and changes of the ground surface due
to the time gap between acquisition of the imagery. A triplet is
constructed out of an along track Ikonos stereo pair taken in
March 2002 and a third image taken in May 2005. The third
image can be considered as a nadir image. Selection criteria for
the near vertical image were multiple: overlap with stereo
couple, cloud-free acquisition, minimal time interval and
optimal stereo constellation. Despite the big time interval, the
2005 Ikonos image was chosen to be the most optimal
candidate.
N
Meters
\ 0 4000
Figure 1. High resolution study field, indicated by the red
polygon. The extent of the HR study area covers the
overlapping area between the 3 Ikonos images.
The Ikonos STEREO product imagery, which comprises of a
forward and backward image acquisition and the GEO Ortho
Kit 2005 image are panchromatic, resampled to a spatial
resolution of 1 m by the image provider and provided with the
Rational Polynomial Coefficients (RPC) camera model file.
Further characteristics of each image of the triplet can be found
in table 1.
Image ID
Acquisition
date
Elevation
angle
Collection
azimuth
Sun
elevation
angle
A (Forward)
1/03/2002
67.59°
1.6°
39.1°
B (Backward)
1/03/2002
75.59°
214.1°
39.1°
C (Nadir)
16/05/2005
80.93°
23.5°
65.5°
Table 1. Characteristics of the three VHR satellite images
acquired over the study field.
Part of the mega city Istanbul, Turkey is chosen as test field for
the project, mainly because it is a city characterized by an
intense urban growth. The city is very compact and
concentrated along the Bosphorus strait. The high resolution
test area covers the overlapping area between the Ikonos 2002
stereo pair and the 2005 image and covers an area of
approximately 60 km 2 , containing Istanbul’s historic peninsula
and going up to the north to the urban fringe. It concerns a
densely built-up area with a height range of 220 m with the
lowest point at sea level and geo-morphologically characterized
by a hilly landscape.
4. SURFACE MODEL GENERATION
In following subsections, the successive steps of the applied
methodology for city surface model generation, based on
(tri)stereoscopic VHR satellite imagery, are elucidated. The
emphasis is especially laid on those phases were research is
done to cope with the complexity of an urban environment.
4.1 Tri-stereoscopic approach
Instead of the standard stereo mapping with two images a tri-
stereoscopic approach is followed. Generation of a DSM using
more than two overlapping images has some interesting
characteristics. First of all, this approach strengthens the image
orientation because of the redundancy in the geometric
reconstruction. Points in object space can be calculated by the
best fit of N convergent image rays instead of two. Secondly the
redundancy leads to a more robust matching, as mismatches and
a unique solution, in case of multiple matching candidates, can
be easier identified. In the stereo case, an object point cannot be
matched if it is located in an occluded area on one or both
images. In the tri-stereoscopic case, the third image is taken
from a different viewing angle. Consequently this leads to a
shift of the occluded areas in the image and enlarges the chance
of a successful match.
Processing of the Ikonos triplet is mainly done with a
photogrammetric software platform, called SAT-PP. SAT-PP is
able to perform image matching on more than two images
simultaneously (Zhang & Gruen, 2006). This is in contrast to
most photogrammetric software packages that are only able to
match two images at the same time.
90
